Localizer meets diagnostic scan: 3D MRA slice planning for a fast, ECG-free, fully automated cardiac exam

Background: Workflow automation for cardiac MR scans aims for speed, to reduce operator dependence and required expertise. Existing methods [1-3] typically employ iterative 2D localizers (at least one per required view), often omitting right heart views or valve planes. This study evaluates a 3D MRA that serves both as a diagnostic scan and localizer for fully automated, ECG-free slice planning, enabling a faster and more comprehensive CMR exam.

Methods: A fully automated cardiac exam was set up using research software packages for AutoPositioning [4] and 3D MRA-based slice planning [5] using a (1.8mm)³ MRA variant of [6]. All cardiac views were then scanned as either compressed sensing cine (left- and right-ventricular long axis, short axis, short axis subset, LVOT, RVOT) or flow (mitral, aortic, pulmonary and tricuspid). For easy and fast preparation, BioMatrix Beat Sensor (Siemens Healthineers, Forchheim, Germany) [7] instead of ECG triggering was used. Figure 1 shows an overview of the exam and multi-planar reformats (MPRs) of the 3D MRA in all detected cardiac views.

Twelve volunteers (8 female, age 48±8) were scanned on two 1.5T MRI systems (MAGNETOM Sola, Siemens Healthineers, Forchheim, Germany). Diagnostic image quality of the 3D MRA was scored (yes/no) for coverage of the whole heart, visibility of the origins of the left main (LM) and right coronary artery (RCA). Slice planning was scored on a 3-point Likert scale (2: Excellent, 1: Acceptable, 0: Incorrect). To evaluate consistency of aortic/pulmonary flow planes, Qp/Qs was measured with cvi42 (Circle, Calgary, Canada).

Results: From the start of the exam, it took 4:36±0:55 min until availability of MRA/slice planning results, 5:02±1:00/11:07±1:26/14:06±2:08 min until morphology/cine/flow scans were finished. Coverage of the whole heart was sufficient in all cases. The origin of the LM was visible in 10/12 cases, RCA origin in 11/12 cases . Slice planning scores are shown in Table 1. Views scored as “Incorrect” on cine/flow images were compared to the MPR output of the corresponding view to distinguish actual planning errors from inconsistent breath-holds. This is visualized in Figure 2, also showing case examples with “Acceptable” scores. Mean Qp/Qs was 0.97±0.085 (n=10, 2 excluded due to flow aliasing).

Conclusion: 3D MRA-based planning showed consistently high planning scores with no misdetections on MPRs. Incorrect planes in cine/flow were likely due to inconsistent breath-holds. Coronary origins and large vessels were visualized in most cases, adding diagnostic value. Total scan time could be further reduced by acquiring multiple cines in one breath-hold, though this was not applied here to allow assessment of field-of-view for each individual plane. This workflow is readily extensible to other contrasts (mapping, LGE, etc.). Overall, the method supports fast, ECG-free, fully automated CMR exams with diagnostic-quality 3D MRA as both scan and localizer.

Table 1: Expert scoring of cardiac views on a 3-point Likert scale (2: Excellent, 1: Acceptable, 0: Incorrect). Scores were first given on cine images acquired in each cardiac view, taking into consideration angulation as well as field-of-view. For views scored as "Incorrect", the corresponding MPR of the cardiac view in the 3D MRA localizer was scored on the same scale to differentiate actual detection failures vs. different respiratory positions between 3D MRA and cine. The short-axis stack was scored separately for the aspects of sufficient slice coverage and angulation/field-of-view.

Cardiac View	Scores on Cine/Flow			Incorrect Cine Reassessed on MPR
	Excellent	Acceptable	Incorrect	Excellent	Acceptable	Incorrect
LV 4CH	6	2	4	4	0	0
LV 2CH	12	0	0	-	-	-
LV 3CH	9	2	1	1	0	0
LVOT	12	0	0	-	-	-
RV 2CH	10	1	1	1	0	0
RV 3CH	10	0	2	2	0	0
RVOT	12	0	0	-	-	-
SAX Slice Coverage	10	2	0	-	-	-
SAX Angulation/FoV	12	0	0	-	-	-
SAX Subset	12	0	0	-	-	-
Mitral Valve	9	3	0	-	-	-
Aortic Valve	8	4	0	-	-	-
Above Aortic Valve	11	1	0	-	-	-
Tricuspid Valve	5	6	1	1	0	0
Pulmonary Flow	6	5	1	1	0	0

Figure 1: (a) Timeline of the fully automated CMR exam, showing acquisition and reconstruction times as well as exemplary results for one volunteer. The LVOT, RVOT, short-axis subset and above aortic valve planes were omitted for space but can be seen in the MPR views. (b) MPRs of the 3D MRA in the detected cardiac orientations. Intersection lines on corresponding views allow confirmation of correct positioning and coverage. For regions outside the field-of-view of the 3D MRA, low-resolution data from the prescan is displayed to allow automated field-of-view computation for the detected cardiac views, the results of which are shown as boxes around each view.
Figure 2: Examples of low scores. (a) Cases where an “Incorrect (0)” score was given for the cine, but the corresponding MPR was rated “Excellent (2)”, indicating a breath-hold mismatch rather than a misdetection of the AI algorithm. (b) Cases where an “Acceptable (1)” score was given, either due to suboptimal angulation (mitral, tricuspid, pulmonary) or too low position of the aortic valve.